We propose to pool data from five HIV cohort studies to answer pivotal questions in the management of patients with HIV infection: when to initiate Highly Active Antiretroviral Therapy (HAART)?, which HAART regimes work better as initial treatment?, and when to switch to another HAART regime? Observational cohort studies are appropriate for this analyses as they are more cost-efficient than randomized trials, use less restrictive eligibility criteria than randomized trials, and, when analyzed appropriately analyzed have provided estimates of the efficacy of HAART close to those obtained from the randomized trials. Nevertheless, because of sample size considerations, it is unlikely that any single cohort will be sufficient to obtain stable estimates. We therefore pool data from five ongoing prospective cohort studies in Western Europe and North America: the Veterans Aging Cohort Study (USA), the Swiss HIV Cohort Study (Switzerland), the French Hospital Database HIV Study (France), GEMES (Spain), and PISCIS (Catalonia and Balearic Islands, Spain). Together, these studies have recruited more than 140,000 HIV infected patients. We will use the data from several observational cohorts in a way that mimics the design of randomized trials as closely as possible. We will then apply inverse probability weighting, g-estimation, and the parametric g- formula (methods developed by Robins and collaborators over the last two decades). These methods are not only theoretically valid to address these questions but have also successfully reproduced results from previous randomized trials when applied to observational data. Our study will inform evidence-based guidelines, and the planning of clinical trials. In addition, the successful application of this advanced methodology as proposed will facilitate understanding and training in causal modeling across leading HIV observational research groups in the United States and Europe.